From 57b49da5e4932327a4576be3901a3cb08c81bae8 Mon Sep 17 00:00:00 2001 From: Tong Li Date: Thu, 13 Mar 2025 16:52:15 +0800 Subject: [PATCH] setup update --- applications/ColossalChat/rl_example.py | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/applications/ColossalChat/rl_example.py b/applications/ColossalChat/rl_example.py index 3d4b8a575..30d56c90b 100644 --- a/applications/ColossalChat/rl_example.py +++ b/applications/ColossalChat/rl_example.py @@ -10,12 +10,12 @@ if __name__ == "__main__": parser.add_argument("-d", "--dataset", type=str, default="data.jsonl") parser.add_argument("-t", "--num-trainers", type=int, default=2) parser.add_argument("-i", "--num-inferencer", type=int, default=2) - parser.add_argument("-ibs", "--inference-batch-size", type=int, default=64) + parser.add_argument("-ibs", "--inference-batch-size", type=int, default=32) parser.add_argument("-imbs", "--inference-microbatch-size", type=int, default=16) parser.add_argument("-tbs", "--train-batch-size", type=int, default=16) - parser.add_argument("-tmbs", "--train-microbatch-size", type=int, default=2) + parser.add_argument("-tmbs", "--train-microbatch-size", type=int, default=1) parser.add_argument("-b", "--backend", type=str, default="transformers") - parser.add_argument("-a", "--algo", type=str, default="GRPO", choices=["Simple, GPRO"]) + parser.add_argument("-a", "--algo", type=str, default="GRPO", choices=["Simple, GRPO"]) args = parser.parse_args() ray.init(address="local", namespace="ray-example")